Acceleration sensor output value correction method
By analyzing and modeling historical data from accelerometers and combining various algorithms to construct a correction method, the problem of misjudgment caused by sensor output data drift was solved, achieving accurate correction and prediction, and improving the reliability of equipment health management.
Patent Information
- Application Number
- CN202610032891.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-12
- Publication Date
- 2026-02-27
AI Technical Summary
During long-term continuous operation, accelerometers are prone to data drift due to factors such as fluctuations in ambient temperature and humidity, mechanical vibration and wear, and aging of circuit components. This can lead to misjudgments in equipment fault diagnosis models and distortions in health status assessments, and may even cause industrial production accidents or loss of control of intelligent systems.
An accelerometer output value correction method is adopted. By acquiring historical operational big data and analyzing data drift trends, algorithms such as linear regression, FCM fuzzy clustering, SVM binary classification and decision tree are used to construct a decision classification tree, predict the fault development type and generate correction values. Combined with Bayesian network to screen effective data, divide the differential drift interval and establish a set of correction algorithms to achieve dynamic correction.
It achieves precise correction of the output value of the accelerometer, avoiding the problems of over- or under-correction in traditional methods, improving recognition accuracy and computational efficiency, and realizing a technological upgrade from post-repair to pre-correction.
Smart Images
Figure CN121577926A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application relates to the technical field of data processing, in particular to an acceleration acceleration sensor output value correction method. BACKGROUND
[0002] The acceleration acceleration sensor is an acceleration sensor capable of measuring acceleration. It is usually composed of a mass block, a damper, an elastic element, a sensitive element and an adaptive circuit, etc. During the acceleration process, the acceleration sensor measures the inertia force acting on the mass block and obtains the acceleration value by using Newton's second law.
[0003] During long-term continuous operation, the acceleration sensor is affected by multiple factors such as environmental temperature and humidity fluctuations, mechanical vibration and wear, and circuit element aging. Its output data is prone to data drift, which is manifested as the deviation between the measured value and the true acceleration value increasing with time. If this drift is not corrected in time, it will lead to misjudgment of the equipment fault diagnosis model, distortion of the health status evaluation, and even cause industrial production accidents or intelligent system out of control, which seriously restricts the application of the equipment health management technology based on acceleration sensor data. SUMMARY
[0004] To solve the above technical problems, the application provides an acceleration acceleration sensor output value correction method, which solves the above problems.
[0005] To achieve the above purposes, the technical scheme adopted by the application is as follows: An acceleration acceleration sensor output value correction method, comprising the following steps: S1, obtaining historical operation big data of a plurality of known acceleration sensors, analyzing the data drift trend of the historical operation big data, generating a data drift trend vector of the plurality of known acceleration sensors, and clustering and dividing according to the drift fault types of the known acceleration sensors.
[0006] S2, based on the fault development type state distribution of the plurality of known acceleration sensors, dividing the real-time operation data of the target acceleration sensor, obtaining the real-time operation data of the target acceleration sensor pointing to the fault development type, marking the support factors of the real-time operation data of the target acceleration sensor pointing to the fault development type according to the cause-effect analysis, and determining the real-time operation data of the target acceleration sensor pointing to the fault development type characteristic data set; S3, according to the failure development type of the known plurality of acceleration sensors, constructing a decision classification tree, predicting the hardware drift state of each stage of the failure development type of the known plurality of acceleration sensors, taking the real-time running data of the target acceleration sensor pointing to the failure development type feature data set as input, generating the hardware drift state of the real-time running data of the target acceleration sensor pointing to the failure development type attribution stage, marking the hardware drift interval, and generating the acceleration sensor output value correction value; Preferably, the step S1 specifically comprises: Based on the historical running big data of the known plurality of acceleration sensors, using criterion detection and rejecting abnormal values in the historical running big data of the known plurality of acceleration sensors, removing invalid time period data, and using linear interpolation to complete missing values; According to the historical running big data of the known plurality of acceleration sensors, unifying the time granularity, and time aligning the historical running big data of the known plurality of acceleration sensors; Based on linear regression, constructing an acceleration sensor drift trend analysis model; According to the acceleration sensor drift trend analysis model, taking the historical running data of unit time as the independent variable and the future measurement value of the known plurality of acceleration sensors as the dependent variable, analyzing the data drift trend of the historical running big data, and generating a data drift trend vector of the known plurality of acceleration sensors; Preferably, the step S1 further comprises: According to the data fitting boundary of the acceleration sensor drift trend analysis model, marking the data points in the data drift trend vector of the known plurality of acceleration sensors that are in the non-fitting region as the running offset feature data of the acceleration sensor; Based on the FCM fuzzy clustering algorithm, constructing an initial clustering cluster based on the drift failure type of the known acceleration sensor, calculating the Euclidean distance and membership of each running offset feature data with respect to each clustering center for the running offset feature data of the acceleration sensor, and generating a membership matrix of the running offset feature data of the acceleration sensor to the drift failure type cluster of the known acceleration sensor; According to the membership matrix, determining the drift failure type to which the running offset feature data of the acceleration sensor belongs, and obtaining the preliminary failure classification result of the running offset feature data of the acceleration sensor; Preferably, the step S2 specifically comprises: Based on the failure development type state distribution of the known plurality of acceleration sensors, extracting the corresponding labels at different development stages of the failure, and obtaining a failure label data vector; The SVM support vector machine is used to take the membership matrix of the known acceleration sensor drift fault type cluster as input, with the fault label data vector and the running offset feature data of the acceleration sensor as input, to obtain the fault development type pointed by the running offset feature data of the known acceleration sensor, and according to the preliminary fault classification result of the running offset feature data of the acceleration sensor, the fault development type is binary classified, and a fault development type binary classification model is constructed. The real-time running data of the target acceleration sensor is collected, and the fault development type pointed by the real-time running data of the target acceleration sensor is predicted according to the fault development type binary classification model. Preferably, the step S2 further comprises: Based on the Bayesian network, the fault development type pointed by the real-time running data of the target acceleration sensor is taken as a target node, and support factors in the fault development type pointed by the real-time running data of the target acceleration sensor are taken as evidence nodes, to construct a Bayesian network topology structure; The prior probability of the fault development type pointed by the real-time running data of the target acceleration sensor is calculated by using a sliding window statistical unit time to count the occurrence frequency of the fault development type pointed by the real-time running data of the target acceleration sensor; According to the statistical calculation of the acceleration sensor historical fault data, the state probability of the fault development type under each support factor is calculated, and a conditional probability table is constructed; The prior probability of the fault development type pointed by the real-time running data of the target acceleration sensor and the conditional probability table are taken as input, and the probability of the support factor in the fault development type pointed by the real-time running data of the target acceleration sensor is calculated by using the Bayesian formula under the condition that the real-time running data of the target acceleration sensor points to the fault development type state, the support factor in the fault development type pointed by the real-time running data of the target acceleration sensor is marked, and the real-time running data of the target acceleration sensor is determined to point to the fault development type feature data set; Preferably, the step S3 specifically comprises: Based on the decision classification tree, the fault development type of the known acceleration sensor is taken as a root node, and each stage of the known fault development type is taken as a branch node, the entropy of the parent node is calculated based on the information gain criterion, the running offset feature data of the known acceleration sensor is split according to the value of each candidate hardware fault distinguishing feature, the weighted sum of the child node entropy is calculated, the difference value between the entropy of the parent node and the weighted sum of the child node entropy is used as the information gain, and the candidate hardware fault distinguishing feature with the maximum information gain is selected as the branch decision as the target, to construct a fault type and stage determination model; According to the fault type and stage determination model, the fault type stage pre-judgment result of the acceleration sensor is obtained, the fault development type, the belonging stage and the corresponding hardware drift state of the known acceleration sensor are output through the model, and the hardware drift state of each stage of the fault development type of the known acceleration sensor is predicted; Preferably, the step S3 further comprises: Based on the statistical analysis method, according to the hardware drift state of each stage of the fault development type of the known acceleration sensor, the data drift characteristic threshold of the corresponding hardware of each stage of the known drift fault development type is counted, the data drift interval of the corresponding hardware of each stage is determined, the corresponding correction logic is implemented for each drift fault type and each drift stage, and a correction algorithm set is established; The real-time running data of the target acceleration sensor is input to the fault development type characteristic data set, the hardware drift state of the real-time running data of the target acceleration sensor is generated according to the fault type and stage determination model, and the hardware drift interval matching the current drift value is marked; According to the drift interval and the corresponding fault type marked by the real-time running data of the target acceleration sensor, the corresponding correction algorithm is determined based on the correction algorithm set, the correction value of the real-time output value of the target acceleration sensor is calculated, and the corrected data is output.
[0007] Compared with the prior art, the beneficial effects of the present application are as follows: 1. Linear regression drift trend modeling, FCM fuzzy clustering fault rough classification, SVM binary classification refined fault development type and decision tree stage determination are combined, which breaks through the limitations of traditional single algorithm in low recognition accuracy of drift fault and fuzzy stage division, and realizes full-link accurate positioning from drift trend capture to fault type and development stage.
[0008] 2. A Bayesian network is constructed by taking the fault development type as a target node and the supporting factors as evidence nodes, the influence weight of the supporting factors is quantified by combining the prior probability and the conditional probability table, adaptive filtering of the feature data set of the target sensor is realized, invalid data interference is avoided, and the input quality and operation efficiency of the subsequent correction model are improved.
[0009] 3. Based on the statistical characteristic threshold of the hardware drift state of each stage of the fault development, the differential drift interval is divided and the corresponding correction algorithm set is established, the customized correction strategy is implemented for different fault types and different development stages, the problems of over-correction or under-correction caused by the traditional correction method are solved, the drift trend and fault type benchmark library is constructed relying on the historical running big data of the known sensor, and the dynamic matching of the fault development type and stage is completed combined with the real-time data of the target sensor, the correction value is generated in the stage when the hardware drift has not caused the fault, and the technical upgrade from post-repair to pre-correction is realized. BRIEF DESCRIPTION OF DRAWINGS
[0010] Figure 1 A schematic diagram of the accelerometer output value correction method; Figure 2 A flowchart illustrating the clustering process for known drift fault types of accelerometers; Figure 3 A schematic diagram illustrating the process of generating a dataset of fault development type characteristics based on real-time running data; Figure 4 This is a schematic diagram of the output process for correcting the output value of the accelerometer. Detailed Implementation
[0011] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.
[0012] Reference Figure 1 As shown, an accelerometer output value correction method includes the following steps: S1. Obtain historical operational big data from several known accelerometers, analyze the data drift trend of the historical operational big data, generate data drift trend vectors for several known accelerometers, and cluster them according to the drift fault types of the known accelerometers. S2. Based on the known state distribution of fault development types of several acceleration sensors, the real-time operating data of the target acceleration sensor is divided to obtain the fault development type pointed to by the real-time operating data of the target acceleration sensor. According to causal analysis, the supporting factors in the fault development type pointed to by the real-time operating data of the target acceleration sensor are marked, and the characteristic dataset of the fault development type pointed to by the real-time operating data of the target acceleration sensor is determined. S3. Based on the known fault development types of several accelerometers, construct a decision classification tree to predict the hardware drift state of each stage of the known fault development types of several accelerometers. Using the real-time operating data of the target accelerometer pointing to the fault development type feature dataset as input, generate the hardware drift state of the stage to which the real-time operating data of the target accelerometer belongs to the fault development type, mark the hardware drift interval, and generate the accelerometer output value correction value.
[0013] Reference Figure 2 As shown, step S1 specifically includes: Based on the historical operational big data of several known accelerometers, outliers in the historical operational big data of several known accelerometers are detected and removed using criteria, invalid time period data are removed, and missing values are filled in using linear interpolation. According to historical operation big data of the known plurality of acceleration sensors, unify the time granularity, and time-align the historical operation big data of the known plurality of acceleration sensors; Based on linear regression, an acceleration sensor drift trend analysis model is constructed. According to the acceleration sensor drift trend analysis model, historical operation data per unit time is taken as an independent variable, and future measurement values of the known plurality of acceleration sensors are taken as a dependent variable, so as to analyze the data drift trend of the historical operation big data and generate a data drift trend vector of the known plurality of acceleration sensors.
[0014] Further, step S1 further includes: According to the data fitting boundary of the acceleration sensor drift trend analysis model, data points in the data drift trend vector of the known plurality of acceleration sensors that are in a non-fitting region are marked as operation offset feature data of the acceleration sensor; Based on the FCM fuzzy clustering algorithm, an initial clustering cluster is constructed according to the drift fault types of the known acceleration sensors, the Euclidean distance and membership of each operation offset feature data relative to each clustering center are calculated for the operation offset feature data of the acceleration sensor, and a membership matrix of the operation offset feature data of the acceleration sensor to the drift fault type cluster of the known acceleration sensor is generated. According to the membership matrix, the drift fault type to which the operation offset feature data of the acceleration sensor belongs is determined, and a preliminary fault classification result of the operation offset feature data of the acceleration sensor is obtained.
[0015] Referring to Figure 3 As shown in the figure, step S2 specifically includes: Based on the fault development type state distribution of the known plurality of acceleration sensors, corresponding labels at different development stages of the fault are extracted, and a fault label data vector is obtained. Using the SVM support vector machine, the fault label data vector and the membership matrix of the operation offset feature data of the acceleration sensor to the drift fault type cluster of the known acceleration sensor are taken as inputs, so as to obtain the fault development type to which the operation offset feature data of the known acceleration sensor points, and according to the preliminary fault classification result of the operation offset feature data of the acceleration sensor, the fault development type is two-classified, and a fault development type two-classification model is constructed. Real-time operation data of the target acceleration sensor is collected, and the fault development type to which the real-time operation data of the target acceleration sensor points is predicted according to the fault development type two-classification model.
[0016] Further, step S2 further includes: Based on the Bayesian network, the real-time running data of the target acceleration sensor is taken as a target node, and the support factors in the real-time running data of the target acceleration sensor are taken as evidence nodes, to construct a Bayesian network topology structure; The prior probability of the real-time running data of the target acceleration sensor being directed to the fault development type is calculated by using a sliding window statistical unit time to count the occurrence frequency of the real-time running data of the target acceleration sensor being directed to the fault development type; According to the historical fault data of the acceleration sensor, the state probability of the fault development type under each support factor is calculated, and a conditional probability table is constructed; The prior probability of the real-time running data of the target acceleration sensor being directed to the fault development type and the conditional probability table are taken as inputs, and the probability of the support factor in the real-time running data of the target acceleration sensor being directed to the fault development type is calculated by using the Bayesian formula under the condition that the real-time running data of the target acceleration sensor is directed to the fault development type, the support factor of the real-time running data of the target acceleration sensor being directed to the fault development type is marked, and the real-time running data of the target acceleration sensor is directed to the fault development type characteristic data set is determined.
[0017] Reference Figure 4 The step S3 specifically includes: Based on the decision classification tree, the fault development type of the known acceleration sensor is taken as a root node, each stage of the known fault development type is taken as a branch node, the entropy of the parent node is calculated based on the information gain criterion, the known acceleration sensor running offset characteristic data is split according to the value of each candidate hardware fault distinguishing feature, the weighted sum of the child node entropy is calculated, the difference between the entropy of the parent node and the weighted sum of the child node entropy is taken as the information gain, the candidate hardware fault distinguishing feature with the maximum information gain is selected as the branch decision as the target, and a fault type and stage determination model is constructed. According to the fault type and stage determination model, the fault type stage prediction result of the acceleration sensor is obtained, the fault development type, the belonging stage and the corresponding hardware drift state of the known acceleration sensor are output through the model, and the hardware drift state of each stage of the fault development type of the known acceleration sensor is predicted.
[0018] Further, the step S3 further includes: Based on the statistical analysis method, the data drift feature threshold of the corresponding hardware of each stage of the known drift fault development type is counted, the data drift interval of the corresponding hardware of each stage is determined, the corresponding correction logic is implemented for each drift fault type and each drift stage, and a correction algorithm set is established. The real-time running data of the target acceleration sensor points to the fault development type feature data set as input, and according to the fault type and stage determination model, the real-time running data of the target acceleration sensor points to the hardware drift state of the fault development type belonging stage, and the current drift value matching hardware drift interval is marked; According to the drift interval and corresponding fault type marked by the real-time running data of the target acceleration sensor, the corresponding correction algorithm is determined based on the correction algorithm set, the correction value of the real-time output value of the target acceleration sensor is calculated, and the corrected data is output.
[0019] The use process of the application is: obtaining historical running big data of known acceleration sensors, analyzing the data drift trend of the historical running big data, generating the data drift trend vector of the known acceleration sensors, and clustering and dividing according to the drift fault type of the known acceleration sensors; according to the fault development type state distribution of the known acceleration sensors, the real-time running data of the target acceleration sensor is divided to obtain the real-time running data of the target acceleration sensor pointing to the fault development type, the supporting factors of the real-time running data of the target acceleration sensor pointing to the fault development type are marked according to the cause-effect analysis, the real-time running data of the target acceleration sensor pointing to the fault development type feature data set is determined; according to the fault development type of the known acceleration sensors, a decision classification tree is constructed, the hardware drift state of each stage of the fault development type of the known acceleration sensors is predicted, the real-time running data of the target acceleration sensor points to the fault development type feature data set as input, the real-time running data of the target acceleration sensor points to the hardware drift state of the fault development type belonging stage, and the hardware drift interval is marked, and the acceleration sensor output value correction value is generated.
[0020] In summary, the application has the advantages that: the linear regression drift trend modeling, FCM fuzzy clustering fault rough classification, SVM binary classification fault development type refinement and decision tree stage judgment break through the limitations of traditional single algorithm low recognition accuracy of drift fault and fuzzy stage division, realize the whole link accurate positioning from drift trend capture to fault type and development stage; the Bayesian network is constructed with fault development type as target node and support factors as evidence node, the influence weight of support factors is quantified combining prior probability and conditional probability table, the adaptive filtering of target sensor feature data set is realized, invalid data interference is avoided, and the input quality and operation efficiency of subsequent correction model are improved; based on the hardware drift state statistical feature threshold of each stage of fault development, the differential drift interval is divided and the corresponding correction algorithm set is established, the customized correction strategy is implemented for different fault types and different development stages, the over-correction or under-correction problem caused by traditional correction method "one size fits all" is solved, the drift trend and fault type benchmark library is constructed relying on known sensor historical operation big data, the dynamic matching of fault development type and stage is completed combining real-time data of target sensor, the correction value can be generated in the stage when hardware drift has not caused fault, and the technical upgrade from post-repair to pre-correction is realized.
[0021] The basic principles, main features and advantages of the present application are shown and described above. It should be understood by those skilled in the art that the present application is not limited by the above examples, and the above examples and descriptions in the specification are only the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection required by the present application is defined by the appended claims and their equivalents.
Claims
1. A method for correcting the output value of an accelerometer, characterized in that, Includes the following steps: S1. Obtain historical operational big data from several known accelerometers, analyze the data drift trend of the historical operational big data, generate data drift trend vectors for several known accelerometers, and cluster them according to the drift fault types of the known accelerometers. S2. Based on the known state distribution of fault development types of several acceleration sensors, the real-time operating data of the target acceleration sensor is divided to obtain the fault development type pointed to by the real-time operating data of the target acceleration sensor. According to causal analysis, the supporting factors in the fault development type pointed to by the real-time operating data of the target acceleration sensor are marked, and the characteristic dataset of the fault development type pointed to by the real-time operating data of the target acceleration sensor is determined. S3. Based on the known fault development types of several accelerometers, construct a decision classification tree to predict the hardware drift state of each stage of the known fault development types of several accelerometers. Using the real-time operating data of the target accelerometer pointing to the fault development type feature dataset as input, generate the hardware drift state of the stage to which the real-time operating data of the target accelerometer belongs to the fault development type, mark the hardware drift interval, and generate the accelerometer output value correction value.
2. The method for correcting the output value of an accelerometer according to claim 1, characterized in that, Step S1 specifically includes: Based on the historical operational big data of several known accelerometer sensors, utilizing... The criteria are used to detect and remove outliers from the historical operational big data of several known accelerometers, eliminate invalid time periods, and use linear interpolation to fill in missing values. Based on the historical operational big data of several known accelerometers, the time granularity is unified, and the historical operational big data of several known accelerometers is time-aligned. A drift trend analysis model for accelerometers is constructed based on linear regression. Based on the accelerometer drift trend analysis model, using historical operating data per unit time as the independent variable and known future measurement values of several accelerometers as the dependent variable, the data drift trend of historical operating big data is analyzed, and a data drift trend vector of several known accelerometers is generated.
3. The method for correcting the output value of an accelerometer according to claim 2, characterized in that, Step S1 further includes: Based on the data fitting boundary of the accelerometer drift trend analysis model, the data points in the non-fit region of the known data drift trend vectors of several accelerometers are marked and denoted as the running offset characteristic data of the accelerometer. Based on the FCM fuzzy clustering algorithm, an initial cluster is constructed with the known drift fault types of the accelerometer. For the operational offset feature data of the accelerometer, the Euclidean distance and membership degree of each operational offset feature data relative to each cluster center are calculated, and the membership degree matrix of the operational offset feature data of the accelerometer to the known drift fault type clusters of the accelerometer is generated. Based on the membership matrix, the drift fault type to which the operational offset feature data of the accelerometer belongs is determined, and the preliminary fault classification results of the operational offset feature data of the accelerometer are obtained.
4. The method for correcting the output value of an accelerometer according to claim 3, characterized in that, Step S2 specifically includes: Based on the known fault development type state distribution of several acceleration sensors, the corresponding labels at different fault development stages are extracted to obtain fault label data vectors. Using SVM (Support Vector Machine), the membership matrix of known drift fault type clusters of accelerometers is obtained by taking fault label data vectors and accelerometer operational offset feature data as inputs. The fault development type pointed to by the known accelerometer operational offset feature data is obtained. Based on the preliminary fault classification results of the accelerometer operational offset feature data, the fault development type is classified into two categories to construct a fault development type binary classification model. Real-time operating data of the target accelerometer is collected, and the fault development type is predicted based on the fault development type binary classification model.
5. The method for correcting the output value of an accelerometer according to claim 4, characterized in that, Step S2 further includes: Based on Bayesian networks, the fault development type pointed to by the real-time operating data of the target accelerometer is used as the target node, and the supporting factors in the fault development type pointed to by the real-time operating data of the target accelerometer are used as evidence nodes to construct a Bayesian network topology. By using a sliding window to statistically analyze the frequency of occurrence of fault development types indicated by the real-time operating data of the target accelerometer per unit time, the prior probability of the fault development type indicated by the real-time operating data of the target accelerometer is calculated. Based on the historical fault data of the accelerometer, the state probability of the fault development type under each supporting factor is calculated, and a conditional probability table is constructed. Using the prior probability and conditional probability table of the real-time operating data of the target accelerometer pointing to the fault development type as input, and given that the real-time operating data of the target accelerometer points to the fault development type, the probability of the supporting factors in the real-time operating data of the target accelerometer pointing to the fault development type is calculated using Bayes' theorem. The supporting factors in the real-time operating data of the target accelerometer pointing to the fault development type are then labeled, and the feature dataset of the real-time operating data of the target accelerometer pointing to the fault development type is determined.
6. The method for correcting the output value of an accelerometer according to claim 5, characterized in that, Step S3 specifically includes: Based on the decision classification tree, the known fault development type of the accelerometer is used as the root node, and each stage of the known fault development type is used as the branch node. Based on the information gain criterion, the entropy of the parent node is calculated. The known accelerometer running offset feature data is split according to the value of each candidate hardware fault distinguishing feature. The weighted sum of the entropy of the child nodes is calculated. The difference between the entropy of the parent node and the weighted sum of the entropy of the child nodes is used as the information gain. With the goal of maximizing the information gain, the candidate hardware fault distinguishing feature with the largest information gain is selected as the branch decision to construct the fault type and stage determination model. Based on the fault type and stage judgment model, the fault type and stage prediction results of the accelerometer are obtained. The model outputs the fault development type, the stage to which it belongs, and the corresponding hardware drift state of the known accelerometer. The hardware drift state of each stage of the fault development type of several known accelerometers is predicted.
7. The method for correcting the output value of an accelerometer according to claim 6, characterized in that, Step S3 further includes: Based on statistical analysis, according to the hardware drift state of each stage of the known fault development type of several acceleration sensors, the data drift characteristic threshold of the hardware corresponding to each stage of the known drift fault development type is statistically analyzed, the data drift interval of the hardware corresponding to each stage is determined, and corresponding correction logic is implemented for each drift fault type and each drift stage to establish a set of correction algorithms. Using the real-time operating data of the target accelerometer pointing to the fault development type feature dataset as input, and based on the fault type and stage determination model, the hardware drift state of the fault development type belonging to the real-time operating data of the target accelerometer is generated, and the hardware drift interval matching the current drift value is marked. Based on the drift range and corresponding fault type marked by the real-time operating data of the target accelerometer, the corresponding correction algorithm is determined based on the set of correction algorithms, the correction value of the real-time output value of the target accelerometer is calculated, and the corrected data is output.